AI that earns its place
A short reflection after speaking with Adam about pixelmesh, and why LoopCoach should show AI improving quality of life rather than just productivity.
I spoke to Adam today about pixelmesh: where it came from, what inspired it, what has improved, and where it might go next.
The useful bit of the conversation was not a feature list. It was the reminder that good tools usually have a person-shaped reason underneath them. Something annoyed someone enough, helped someone enough, or mattered enough that it became worth building properly.
pixelmesh is a good example because the idea is simple to explain and oddly hard to build well. A room full of people open a URL, hold up their phones, and the audience becomes a single synchronised display. Each phone is still its own device, but together they behave like one canvas.
The cleverness is in the coordination. Every device gets a shared clock and a known position. A camera at the front watches the audience while screens blink optical patterns, then maps the room into a coordinate grid. The site says the current version can do that in about thirteen seconds. From there, the controller sends effect parameters and each device renders its own frame locally.
That is a lovely systems idea because the distributed bit becomes visible. Most distributed systems disappear under the surface. We see the app, not the coordination. pixelmesh makes the coordination the thing you are there to watch.
The improvement path is interesting too. The first public version used printed AprilTag markers for calibration. The current version replaced those with optically encoded screen blinks, so the setup gets closer to the right kind of magic: no install, no scans, no per-device faff. Open a URL, hold up the screen, become part of the display.
That has been sitting with me because it is exactly the bit I want to bring out with LoopCoach.
Most public AI examples still seem to land in three buckets: fun demos, business automation, or content machinery. Some of that is useful. Some of it is noise. But it is not the whole story. The part I keep coming back to is quieter: AI as a way to reduce cognitive load around something that already matters.
For us, that means Type 1 diabetes. Not replacing judgement. Not making medical decisions for a child. Not pretending a model understands a family better than the family does. The valuable version is much more boring and much more useful: keeping context together, spotting patterns, turning raw data into better questions, and giving parents a clearer starting point when they are already tired.
That is the showcase I think LoopCoach needs to become. Not a shiny AI demo. A small example of AI pointed at quality of life.
There is a difference between a tool that impresses you for five minutes and a tool that quietly makes a hard thing a little easier every week. The second kind is harder to demo. It has fewer fireworks. It also matters more.
That is probably the thread I want to pull on next: AI that earns its place by taking weight off real people, not by looking clever in isolation.